Saturday, April 30, 2016

Season 6 superfinal, games 37-40


This is part 7 of my looking back at the Komodo-Stockfish superfinal match of season 6. Previous parts can be found at
part 1, games 1-8
part 2, games 9-16
part 3, games 17-24 
part 4, games 25-28
part 5, games 29-32 
part 6, games 33-36 

After 36 games Stockfish led 7-3 with 26 draws.

In game 37 Stockfish had advanced pawns on the queen side, but its king was vulnerable on the king side. Komodo gave a pawn while keeping the pressure with its queen and knight, seeing potential repetitions or perpetual checks as a probable outcome.


Stockfish kept a small eval advantage and continued to advance pawns on the queen side, ignoring black's attack. Komodo was sure that a repetition has started on move 39, but Stockfish did not go along with this and let its king walk away from the black QN combination.


It turned out that Stockfish was right. Once the threat of perpetual check was gone the white queen side pawns became a big advantage. Eventually Komodo had to give a rook to get rid of the white pawns. The resulting QRB vs QN endgame was not easy but Stockfish managed to get the queens exchanged and the game was soon over.

The reverse game was less eventful. Both kings were safe, there was some action on the queen side due to an open file, but the evals drifted down to 0 after 24 moves and stayed there until a draw was reached.

Game 39 started with a small eval advantage for white. Stockfish used a pawn attack on the king side, exposing its castled king while maneuvering a knight to h6.

Komodo's king and rook were trapped. With the f file open Stockfish applied pressure to f7, and Komodo had less space to move and defend. After a series of exchanges in the center by move 40 the position opened up. Komodo exchanged RP for BN, but the trap in the corner meant Stockfish had a full bishop advantage in the center.


After forcing a queen exchange Stockfish continued to clear some of the pawns. Komodo broke out of the trap by exchanging the rook with the trapping knight. The resulting RB vs R endgame was a clear win for Stockfish.

In the reverse game Komodo did not move its king side pawns and both kings remained safe. By move 24 both evals reached 0, and it was only a matter of time until the game was adjudicated. That's twice in a row that Stockfish had a clear win over Komodo in a pair of reverse games. With 24 games remaining it is very unlikely that Komodo can win this match.

Friday, April 22, 2016

Season 9 is approaching, May 2016


TCEC season 9 is starting soon!!

After a 5 month break Chessdom have indicated that season 9 will start on May 1st 2016, less than two weeks away. From the information released so far this season will feature:

- up to 32 engines (28 confirmed)
- several new engines, some of which are strong (over 3000 ELO)
- a long season, 7 months long
- 15 second increments

If there are better engines this season, will there be a struggle for top places? In season 8 Gull was strong enough to challenge Komodo and Stockfish in stage 3, will there be others this season? Houdini is also being developed again after a long time, can it catch up with the leaders and return to the top as it was a few seasons ago?

Seven months is really long, almost twice as long as all the seasons I've watched so far (see table below). I really hope I can keep watching the games without too much disruption to my life.

  • Season 4: 32 engines, 364 games, 108 days
  • Season 5: 36 engines, 525 games, 98 days
  • Season 6: 36 engines, 590 games, 115 days
  • Season 7: 28 engines, 521 games, 100 days
  • Season 8: 24 engines, 454 games, 102 days
  • Season 9: 32 engines, 804 games, 215 days (planned)
 
Remember Stockfish's losses on time last season? I hope the developers fixed it this time, the increment is going to be only 15 seconds this time.

As in last season, I intend to post a statistics report after each stage is over. These are based on the PGN files in the TCEC archive. Have a look at my season 8 posts to see what I compute, nothing too fancy. If there is an analysis you would like to see, post a comment below. If I think it's interesting and not too difficult I will add it to the report.

Go TCEC!!

Sunday, April 10, 2016

Season 6 superfinal, games 33-36


This is part 6 of my looking back at the Komodo-Stockfish superfinal match of season 6. Previous parts can be found at
part 1, games 1-8
part 2, games 9-16
part 3, games 17-24 
part 4, games 25-28
part 5, games 29-32 

After 32 games Stockfish led 6-2 with 24 draws.

Game 33  began with white  a pawn down but with the black king exposed and the black queen too far away and also vulnerable. As a result Stockfish was able to develop its pieces faster than Komodo, with an eval advantage of 1.3. On move 20 Stockfish purposefully entered a queen-rook fork threat.


After the rook for knight exchange white's attack looked dangerous again. Komodo defended its king, exchanging pieces to remove threats, trading a rook back for a bishop and exchanging pieces. When the dust cleared Stockfish was a pawn up in a RN vs RB game.


Was all this worth it for just a pawn (at least it's a passer...)? Komodo thought it was moderately safe for a while, it even got the pawn back. The game became a pawn race, the white queen side and the black king side.


Stockfish had a crucial head start, Komodo was forced to lose the bishop for one pawn, and the remaining rook could not stop two supported white pawns on its own.

In the reverse game 34 Komodo was more patient as white, allowing Stockfish to exchange queens and castle its king. This led to a quiet position where white had more space and a bishop pair, for black's extra pawn.


Stockfish felt safe but had no plan how to continue, and it basically waited for more than 30 moves while Komodo slowly shifted pieces on the board and the eval slowly increased.


Komodo opened up the center with f5 and a rook exchange, and the evals jumped quickly. The bishop pair became very strong, forcing Stockfish to lose material to get out a mating net. The RB vs R endgame was easily won by Komodo. Again two wins in this opening, but with very different styles.

In game 35 white had to move the king early without castling, but it was not under serious attack. After a series of exchanges only QRR vs QRR remained at move 24. Stockfish was a pawn up but Komodo had an effective attack on the white king. Temporarily white was two pawns up but had to give them back and exchange the rooks to reach safety by move 38. The Q vs Q endgame was drawn after 20 more moves.

The first few moves of game 36 followed the previous game. Komodo however kept its pieces and developed threats on the king side, with eval climbing to about 1. After a series of exchanges the pieces left at move 34 were RRN vs RRB. By move 45 a pair of rooks was exchanged and the eval was falling back to equality. Komodo was able to get a pawn but this only led to a drawn rook ending.

Thursday, April 7, 2016

Season 6 superfinal, games 29-32


This is part 5 of my looking back at the Komodo-Stockfish superfinal match of season 6. Previous parts can be found at
part 1, games 1-8
part 2, games 9-16
part 3, games 17-24 
part 4, games 25-28

After 28 games Stockfish led 5-1 with 22 draws.

Game 29 started with white a pawn down. By move 19 Stockfish gave a rook and another pawn for a bishop, with a strong attack on black through the center. Komodo's king couldn't castle, and after exchanging several pieces it had to give its queen for a rook and bishop in order to get out of the attack. After move 33 only Q vs RR with pawns remained. Stockfish had a small eval advantage but after most pawns were taken Komodo forced a perpetual check for the draw.
In the reverse game 30 Komodo was a pawn down but had a strong initiative and an eval advantage of 1.5, with Stockfish unable to castle and forced to defend. Most pieces remained in play and by move 30 Stockfish built a pawn wall across the board, which appeared very strong.

After a long stretch of shuffle moves with very few pawn moves, and with the 50-move counter down to 10 there was a bishop for knight exchange. The evals remained steady for a few more moves, but then started to increase for no apparent reason. Komodo's advantage became evident when it forced a queen and rook exchange and won two pawns.


 Stockfish could not stop the passer or the white king on the queen side, and Komodo won a few moves later. This was Komodo's first real win of the match, and it came out of what seemed to be a sure draw - the only advantage Komodo has over Stockfish that we've seen so far in this match.

Game 31 started with many exchanges in the center, leaving only QRRB vs QRRB after 20 moves.  Stockfish had a small eval advantage, probably because of Komodo's weak pawns.


Stockfish continued to apply pressure on the black pawns, with eval climbing slowly. After queens were exchanged the evals were at about 1, and there followed a series of moves where Komodo lost a pawn while exchanging most pieces.


Usually a pawn advantage is not enough for a win in rook endings, even two pawns in the f+h (a+c here) case which was possible in this position. However, here Komodo chose to exchange on the g file and Stockfish immediately saw a path to victory. Komodo realized it was losing a few moves later. Was it a forced win for Stockfish or a Komodo blunder?

Game 32 had a similar beginning, with the same pieces remaining on move 20. The difference was that now both engines had a few weak pawns on the queen side. The queens were exchanged quickly, and the rooks a few moves later, by move 28 only B vs B with pawns remained. The kings could not penetrate the pawn lines, and it took more than 50 more moves for the engines to agree that the game was a draw. Stockfish struck back immediately and the score at the halfway point was 6-2 with 24 draws.

Monday, April 4, 2016

Season 6 superfinal, games 25-28


This is part 4 of my looking back at the Komodo-Stockfish superfinal match of season 6. Previous parts can be found at
part 1, games 1-8
part 2, games 9-16
part 3, games 17-24

After 24 games Stockfish led 4-0 with 20 draws.

Game 25 was a quick draw. Stockfish developed an attack on the king side which was not strong enough to get an advantage. The game ended after 32 moves with a bishop sacrifice and perpetual check. In game 26 after the queens were exchanged it seemed that Komodo had some advantage with a pair of bishops to Stockfish's pair of knights. Both evals were 0 though, and indeed Stockfish trapped one of the bishops and exchanged it for a knight. The game continued only until a long enough stretch without pawns or captures triggered the draw rule.

Game 27 started with white a pawn down, followed by opposite direction castling. Stockfish sacrificed a second pawn to open up the king side and attack the black king. Komodo pushed the a-pawn against the white king, and both engines anticipated a rook for bishop exchange that never came.


Instead, after a bishop exchange on move 22 the engines went through a 9-move combination appearing in both PVs, resulting in a QR vs RRNP game.

The white queen was very active, and the black rooks had to defend the back rows. This allowed Stockfish to clear the queen side pawns, gaining two pawns and creating two passers.


Komodo's only threat was the h-pawn, but when the white pawns advanced and the king joined from behind the game was over.
In the reverse game 28 the engines had different plans. Komodo as white was also two pawns down, while Stockfish had an uncastled king and a weak undeveloped queen side.

Stockfish gave one pawn back but then had to lose a knight while its king was struggling in the center.

This was a winning position for Komodo. After queens were exchanged white attacked the queen side, slowly gaining pawns and advancing a passer until a win was declared. Komodo got its first win, but perhaps the opening was too biased - both engines won playing white.